Tag Archives: European Windstorm Model

Europe’s Winter Windstorms – the Only Certainty is Uncertainty

The annual damage from European windstorms can range significantly: from years when there are clusters of severely damaging storms to other years with almost no windstorm loss. How much of this volatility can we predict, and how much remains a roll of the dice? And more specifically, what storm activity can we expect over the next few months?

Forecasting Storminess

Our understanding of the drivers of annual storminess has increased greatly in recent years, allowing us to provide more forecasting insight than ever before. However, there is a cautionary tale for the industry, one that shows the limitations of even the most sophisticated seasonal forecasts.

The Atlantic Multidecadal Oscillation (AMO) is a pattern of long-duration variability in sea surface temperature in the North Atlantic. It is known to influence the climate over much of the northern hemisphere including the level of storminess in Europe1. As north-south gradients of heat in the Atlantic act to fuel extra-tropical storms2 these longer term changes in sea surface temperature tend to alter the odds of extreme storm occurrence over timescales of 60-80 years. Today, the ongoing positive (warm) phase of the AMO favors lower than average storminess this winter.

Annual average values for the AMO Index, 1856-2015 (data from NOAA ESRL3). Positive values (red bars) indicate warmer sea surface temperatures across the North Atlantic, while negative values (blue bars) indicate cooler temperatures.

That’s the multi-decadal perspective. But it will come as no surprise for Europeans to hear that as well as these longer phases of relative activity and inactivity, the continent also experiences variability of storminess from year to year. We know that the jet stream is a main ingredient of storms, and that in turn these storms strengthen the jet itself, in a positive feedback loop that leads to the term “eddy-driven jet.”  This “storms-beget-storms” mechanism typically plays out over a few weeks, and more severe storms are likelier to occur during these periods. The positive feedback between jet and storms amplifies swings in annual damage, and explains a substantial amount of the storm clustering found in longer range historical weather records4. This coupling between storms and jet is reflected in the version 16.0 of the RMS Europe Windstorm Clustering Model.

Researchers have identified various drivers of seasonal storminess in the North Atlantic which, for the coming winter, are ambiguous. For instance: we are three years after the peak of a prolonged but subdued solar cycle and this timing suggests less forcing of storminess. But in contrast the predictions are for neutral to weak La Niña phases of the El Niño–Southern Oscillation (ENSO) which points to a chance of increased forcing of North Atlantic storminess. Whilst, to complicate things further, the anticipated values of tropical stratosphere winds, linked to the Quasi-Biennial Oscillation (QBO), are related to less storminess in the mid-latitude Atlantic – with the caveat that they are in an unusually disrupted pattern.

So is it possible to get off the meteorological fence and make a call? Yes: overall, the multi-decadal and seasonal drivers indicate slightly below average storminess.

Severe Events Can Occur During Any Season

But this does not mean that we as an industry should be entirely relaxed about the new storm season, as the outlook for annual storm damage is blurred by the vagaries of local weather. This is exemplified by storm Kyrill in January 2007.

Then, ahead of the 2006/07 winter, the seasonal and multi-decadal drivers indicated below average storminess, just as they do today. But Kyrill occurred and turned an otherwise innocuous season into a bad one for many. The gusts and damage during this storm were much more extreme than its general circulation, because convection cells embedded in the cold front contributed to extreme damage intensity in some areas5. Storm Kyrill showed how processes on small space and time scales can dominate annual storm damage. These drivers have seriously short predictability windows of just a few hours.

More generally, some of the past variations in annual storminess have no known driver. We are not quite sure how much, but a reasonable ball-park figure is one half. This random part is found in climate models, where the tiniest possible changes at the start of a forecast often grow into large changes in seasonal average storminess.

Although our understanding of the drivers of storminess has greatly increased over the past few years and the odds do favor less storm damage this winter, we should not be complacent. As its tenth anniversary approaches, Storm Kyrill reminds us that major losses can happen in any season, regardless of the forecast.

Web links to references above

1Peings and Magnusdottir (2014)  [ http://iopscience.iop.org/article/10.1088/1748-9326/9/3/034018/pdf ]

2Shaffrey and Sutton (2006)  [ http://journals.ametsoc.org/doi/pdf/10.1175/JCLI3652.1 ]

3NOAA ESRL AMO data [http://www.esrl.noaa.gov/psd/data/timeseries/AMO/ ]

4Cusack (2016)  [ http://www.nat-hazards-earth-syst-sci.net/16/901/2016/nhess-16-901-2016.pdf ]

5Fink et al. (2009)  [ http://centaur.reading.ac.uk/32783/1/nhess-9-405-2009.pdf ]

This post was co-authored by Peter Holland and Stephen Cusack.

European Windstorm: Such A Peculiarly Uncertain Risk for Solvency II

Europe’s windstorm season is upon us. As always, the risk is particularly uncertain, and with Solvency II due smack in the middle of the season, there is greater imperative to really understand the uncertainty surrounding the peril—and manage windstorm risk actively. Business can benefit, too: new modeling tools to explore uncertainty could help (re)insurers to better assess how much risk they can assume, without loading their solvency capital.

Spikes and Lulls

The variability of European windstorm seasons can be seen in the record of the past few years. 2014-15 was quiet until storms Mike and Niklas hit Germany in March 2015, right at the end of the season. Though insured losses were moderate[1], had their tracks been different, losses could have been so much more severe.

In contrast, 2013-14 was busy. The intense rainfall brought by some storms resulted in significant inland flooding, though wind losses overall were moderate, since most storms matured before hitting the UK. The exceptions were Christian (known as St Jude in Britain) and Xaver, both of which dealt large wind losses in the UK. These two storms were outliers during a general lull of European windstorm activity that has lasted about 20 years.

During this quieter period of activity, the average annual European windstorm loss has fallen by roughly 35% in Western Europe, but it is not safe to presume a “new normal” is upon us. Spiky losses like Niklas could occur any year, and maybe in clusters, so it is no time for complacency.

Under Pressure

The unpredictable nature of European windstorm activity clashes with the demands of Solvency II, putting increased pressure on (re)insurance companies to get to grips with model uncertainties. Under the new regime, they must validate modeled losses using historical loss data. Unfortunately, however, companies’ claims records rarely reach back more than twenty years. That is simply too little loss information to validate a European windstorm model, especially given the recent lull, which has left the industry with scant recent claims data. That exacerbates the challenge for companies building their own view based only upon their own claims.

In March we released an updated RMS Europe Windstorm model that reflects both recent and historic wind history. The model includes the most up-to-date long-term historical wind record, going back 50 years, and incorporates improved spatial correlation of hazard across countries together with a enhanced vulnerability regionalization, which is crucial for risk carriers with regional or pan-European portfolios. For Solvency II validation, it also includes an additional view based on storm activity in the past 25 years. Pleasingly, we’re hearing from our clients that the updated model is proving successful for Solvency II validation as well as risk selection and pricing, allowing informed growth in an uncertain market.

Making Sense of Clustering

Windstorm clustering—the tendency for cyclones to arrive one after another, like taxis—is another complication when dealing with Solvency II. It adds to the uncertainties surrounding capital allocations for catastrophic events, especially due to the current lack of detailed understanding of the phenomena and the limited amount of available data. To chip away at the uncertainty, we have been leading industry discussion on European windstorm clustering risk, collecting new observational datasets, and developing new modeling methods. We plan to present a new view on clustering, backed by scientific publications, in 2016. These new insights will inform a forthcoming RMS clustered view, but will be still offered at this stage as an additional view in the model, rather than becoming our reference view of risk. We will continue to research clustering uncertainty, which may lead us to revise our position, should a solid validation of a particular view of risk be achieved.

Ongoing Learning

The scientific community is still learning what drives an active European storm season. Some patterns and correlations are now better understood, but even with powerful analytics and the most complete datasets possible, we still cannot yet forecast season activity. However, our recent model update allows (re)insurers to maintain an up-to-date view, and to gain a deeper comprehension of the variability and uncertainty of managing this challenging peril. That knowledge is key not only to meeting the requirements of Solvency II, but also to increasing risk portfolios without attracting the need for additional capital.

[1] Currently estimated by PERILS at 895m Euro, which aligns with the RMS loss estimate in April 2015

What Is In Store For Europe Windstorm Activity This Winter

From tropical volcanoes to Arctic sea-ice, recent research has discovered a variety of sources of predictability for European winter wind climate. Based on this research, what are the indicators for winter storm damage this season?

The most notable forcings of winds this winter – the solar cycle and the Arctic sea-ice extents – are forcing in opposite directions. We are unsure which forcing will dominate, and the varying amplitude of these drivers over time confuses the situation further: the current solar cycle is much weaker than the past few, and big reductions in sea-ice extent have occurred over the past 20 or so years, as shown in the graph below.

Figure: Standardized anomalies of Arctic sea-ice extent over the past 50 years. (Source: NSIDC)

There are two additional sources of uncertainty, which further undermine predictive skill. First, researchers examine strength of time-mean westerly winds over 3-4 months, whereas storm damage is usually caused by a few, rare days of very strong wind. Second, storms are a chaotic weather process – a chance clash of very cold and warm air – which may happen even when climate drivers of storm activity suggest otherwise.

RMS has performed some preliminary research using storm damages, rather than time-mean westerlies, and we obtain a different picture for East Pacific El Niños. Most of them have elevated storm damage in the earlier half of the storm season (before mid-January) and less later on. Of special note are the two storms Lower Saxony in November 1972 and 87J in October 1987: the biggest autumn storms in the past few decades happened during East Pacific El Niños. The possibility that East Pacific El Niños alter the seasonality of storms, and perhaps raise the chances of very severe autumn storms, highlights potential gaps in our knowledge that compromise predictions.

We have progressed to the stage that reliable, informative forecasts could be issued on some occasions. For instance, large parts of Europe would be advised to prepare for more storm claims in the second winter after an explosive, sulphur-rich, tropical volcano. Especially if a Central Pacific La Niña is occurring [vi] and we are near the solar cycle peak.

However, the storm drivers this coming winter have mixed signals and we dare not issue a forecast. It will be interesting to see if there is more damage before rather than after mid-January, and whatever the outcome, we will have one more data point to improve forecasts of winter storm damage in the future.

Given the uncertainty in windstorm activity levels, any sophisticated catastrophe model should give the user the possibility of exploring different views around storm variability, such as the updated RMS Europe Windstorm Model, released in April this year.

[i] Fischer, E. et al. “European Climate Response to Tropical Volcanic Eruptions over the Last Half Millennium.” Geophys. Res. Lett. Geophysical Research Letters, 2007, .
[ii] Brugnara, Y., et al. “Influence of the Sunspot Cycle on the Northern Hemisphere Wintertime Circulation from Long Upper-air Data Sets.” Atmospheric Chemistry and Physics Atmos. Chem. Phys., 2013.
[iii] Graf, Hans-F., and Davide Zanchettin. “Central Pacific El Niño, the “subtropical Bridge,” and Eurasian Climate.” J. Geophys. Res. Journal of Geophysical Research, 2013.
[iv] Baldwin, M. P., et al. “The Quasi-Biennial Oscillation.” Reviews of Geophysics, 2001.
[v] Budikova, Dagmar. “Role of Arctic Sea Ice in Global Atmospheric Circulation: A Review.” Global and Planetary Change, 2009.
[vi] Zhang, Wenjun, et al. “Impacts of Two Types of La Niña on the NAO during Boreal Winter.” Climate Dynamics, 2014.

Matching Modeled Loss Against Historic Loss in European Windstorm Data

To be Solvency II compliant, re/insurers must validate the models they use, which can include comparisons to historical loss experience. In working towards model validation, companies may find their experience of European windstorm hazard does not match the modeled loss. However, this seeming discrepancy does not necessarily mean something is wrong with the model or with the company’s loss data. The underlying timelines for each dataset may simply differ, which can have a significant influence for a variable peril like European windstorm.

Most re/insurers’ claims records only date back 10 to 20 years, whereas European windstorm models use much longer datasets – generally up to 50 years of the hazard. Looking over the short term, the last 15 years represented a relative lull in windstorm activity, particularly when compared to the more extreme events that occurred in the very active 1980s and 1990s.

Netherlands windstorm variability







RMS has updated its European windstorm model specifically to support Solvency II model validation. The enhanced RMS model includes the RMS reference view, which is based on the most up-to-date, long-term historical record, as well as a new shorter historical dataset that is based on the activity of the last 25 years.

By using the shorter-term view, re/insurers gain a deeper understanding of how historical variability can impact modeled losses. Re/insurers can also perform a like-for-like validation of the model against their loss experience, and develop confidence in the model’s core methodology and data. Alternate views of risk also support a deeper understanding of risk uncertainty, which enhances model validation and provides greater confidence in the models that are used for risk selection and portfolio management.

Beyond Solvency II validation, the model also empowers companies to explore the hazard variability, which is vitally important for a variable peril like European windstorm. If a catastrophe model and a company rely on different but equally valid assumptions, the model can present a different perspective to provide a more complete view of the risk.